Online Recursive Power Management Strategy based on the Reinforcement Learning Algorithm with Cosine Similarity and a Forgetting Factor

In this article, an online recursive power management strategy based on the reinforcement learning technique is proposed to achieve an optimal power distribution and excellent economic performance of a plug-in fuel cell hybrid electric vehicle. First, the optimal control algorithm of the power management strategy is formulated. Then, an online recursive algorithm using cosine similarity and a forgetting factor to achieve the adaptability of the proposed strategy to various driving conditions is formulated. The parameters updating framework is triggered once the change rate of cosine similarity exceeds the threshold value of the corresponding driving cycle. Some of the important parameters such as the learning rate, discount factor, forgetting factor, and the change rate of cosine similarity are analyzed to optimize the online updating capability of the strategy. Finally, an experimental study is conducted to validate the effectiveness of the proposed strategy. Both a rule-based strategy and an equivalent consumption minimization strategy are implemented to establish the benchmark framework used to analyze the performance of the proposed strategy. The simulation and experimental results corroborate the effectiveness and economic performance of the proposed strategy. A comparison with the existing standard benchmark strategy indicates the superior performance of the developed strategy reaching a remarkable reduction in energy consumption at a long driving distance.